Clustering of web users' access patterns using a modified competitive agglomerative algorithm

Web recommendation systems are helpful in overcoming the excess information on web by retrieving the information required by the user with respect to user's or similar users' preferences and interests. In order to make web recommendation system work, web users have to be clustered based on their common interest. The web user clusters are used to obtain the knowledge about the web pages accessed. This knowledge can be used for prefetching of web pages, finding web pages that are frequently accessed together, etc. This paper presents a modification in the original CA clustering algorithm for grouping of web users with respect to access pattern of web pages. The original CA algorithm uses the basic "Fuzzy C Means (FCM) algorithm" to compute the membership matrix. The modified CA algorithm uses a superior FCM algorithm, namely, the Density Weighted FCM (DWFCM) instead of the basic FCM. Experiments are conducted on the datasets obtained from the UCI repository. It is found that the modified CA clustering method exhibits superior capability of clustering when compared to the CA clustering method.